Deep learning-based electroencephalography analysis: a systematic review

Journal of Neural Engineering, Год журнала: 2019, Номер 16(5), С. 051001 - 051001

Опубликована: Май 31, 2019

Abstract Context . Electroencephalography (EEG) is a complex signal and can require several years of training, as well advanced processing feature extraction methodologies to be correctly interpreted. Recently, deep learning (DL) has shown great promise in helping make sense EEG signals due its capacity learn good representations from raw data. Whether DL truly presents advantages compared more traditional approaches, however, remains an open question. Objective In this work, we review 154 papers that apply EEG, published between January 2010 July 2018, spanning different application domains such epilepsy, sleep, brain–computer interfacing, cognitive affective monitoring. We extract trends highlight interesting approaches large body literature order inform future research formulate recommendations. Methods Major databases the fields science engineering were queried identify relevant studies scientific journals, conferences, electronic preprint repositories. Various data items extracted for each study pertaining (1) data, (2) preprocessing methodology, (3) design choices, (4) results, (5) reproducibility experiments. These then analyzed one by uncover trends. Results Our analysis reveals amount used across varies less than ten minutes thousands hours, while number samples seen during training network few dozens millions, depending on how epochs are extracted. Interestingly, saw half publicly available there also been clear shift intra-subject inter-subject over last years. About convolutional neural networks (CNNs), recurrent (RNNs), most often with total 3–10 layers. Moreover, almost one-half trained their models or preprocessed time series. Finally, median gain accuracy baselines was all studies. More importantly, noticed suffer poor reproducibility: majority would hard impossible reproduce given unavailability code. Significance To help community progress share work effectively, provide list recommendations emphasize need reproducible research. our summary table invite authors contribute it directly. A planned follow-up will online public benchmarking portal listing results.

Язык: Английский

---

A Deep Learning Architecture for Temporal Sleep Stage Classification Using Multivariate and Multimodal Time Series

IEEE Transactions on Neural Systems and Rehabilitation Engineering, Год журнала: 2018, Номер 26(4), С. 758 - 769

Опубликована: Март 7, 2018

Sleep stage classification constitutes an important preliminary exam in the diagnosis of sleep disorders. It is traditionally performed by a expert who assigns to each 30 s signal stage, based on visual inspection signals such as electroencephalograms (EEGs), electrooculograms (EOGs), electrocardiograms, and electromyograms (EMGs). We introduce here first deep learning approach for that learns end-to-end without computing spectrograms or extracting handcrafted features, exploits all multivariate multimodal polysomnography (PSG) (EEG, EMG, EOG), can exploit temporal context 30-s window data. For modality, layer linear spatial filters array sensors increase signal-to-noise ratio, last feeds learnt representation softmax classifier. Our model compared alternative automatic approaches convolutional networks decisions trees. Results obtained 61 publicly available PSG records with up 20 EEG channels demonstrate our network architecture yields state-of-the-art performance. study reveals number insights spatiotemporal distribution interest: good tradeoff optimal performance measured balanced accuracy use 6 2 EOG (left right) 3 EMG chin channels. Also exploiting 1 min data before after segment offers strongest improvement when limited are available. As experts, system nature order deliver small computational cost.

Язык: Английский

---

SeqSleepNet: End-to-End Hierarchical Recurrent Neural Network for Sequence-to-Sequence Automatic Sleep Staging

IEEE Transactions on Neural Systems and Rehabilitation Engineering, Год журнала: 2019, Номер 27(3), С. 400 - 410

Опубликована: Янв. 31, 2019

Automatic sleep staging has been often treated as a simple classification problem that aims at determining the label of individual target polysomnography epochs one time. In this paper, we tackle task sequence-to-sequence receives sequence multiple input and classifies all their labels once. For purpose, propose hierarchical recurrent neural network named SeqSleepNet (source code is available http://github.com/pquochuy/SeqSleepNet). At epoch processing level, consists filterbank layer tailored to learn frequency-domain filters for preprocessing an attention-based designed short-term sequential modeling. placed on top learned epoch-wise features long-term modeling epochs. The then carried out output vectors every time step produce labels. Despite being hierarchical, present strategy train in end-to-end fashion. We show proposed outperforms state-of-the-art approaches, achieving overall accuracy, macro F1-score, Cohen's kappa 87.1%, 83.3%, 0.815 publicly dataset with 200 subjects.

Язык: Английский

---

Joint Classification and Prediction CNN Framework for Automatic Sleep Stage Classification

IEEE Transactions on Biomedical Engineering, Год журнала: 2018, Номер 66(5), С. 1285 - 1296

Опубликована: Окт. 22, 2018

Correctly identifying sleep stages is important in diagnosing and treating disorders. This work proposes a joint classification-and-prediction framework based on CNNs for automatic staging, and, subsequently, introduces simple yet efficient CNN architecture to power the framework. Given single input epoch, novel jointly determines its label (classification) neighboring epochs' labels (prediction) contextual output. While proposed orthogonal widely adopted classification schemes, which take one or multiple epochs as inputs produce decision target we demonstrate advantages several ways. First, it leverages dependency among consecutive while surpassing problems experienced with common schemes. Second, even model, has capacity decisions, are essential obtaining good performance ensemble-of-models methods, very little induced computational overhead. Probabilistic aggregation techniques then leverage availability of decisions. We conducted experiments two public datasets: Sleep-EDF Expanded 20 subjects, Montreal Archive Sleep Studies dataset 200 subjects. The yields an overall accuracy 82.3% 83.6%, respectively. also show that not only superior baselines schemes but outperforms existing deep-learning approaches. To our knowledge, this first going beyond standard single-output consider multitask neural networks staging. provides avenues further studies different neural-network architectures

Язык: Английский

---

A Novel Multi-Class EEG-Based Sleep Stage Classification System

IEEE Transactions on Neural Systems and Rehabilitation Engineering, Год журнала: 2017, Номер 26(1), С. 84 - 95

Опубликована: Ноя. 21, 2017

Sleep stage classification is one of the most critical steps in effective diagnosis and treatment sleep-related disorders. Visual inspection undertaken by sleep experts a time-consuming burdensome task. A computer-assisted system thus essential for both disorders monitoring. In this paper, we propose to classify wake stages with high rates sensitivity specificity. The EEG signals 25 subjects suspected sleep-disordered breathing, 20 healthy from three data sets are used. Every epoch decomposed into eight subband epochs each which has frequency band pertaining rhythm (i.e., delta, theta, alpha, sigma, beta 1, 2, gamma or 2). Thirteen features extracted epoch. Therefore, 104 totally obtained every Kruskal–Wallis test used examine significance features. Non-significant discarded. minimal-redundancy-maximal-relevance feature selection algorithm then eliminate redundant irrelevant selected classified random forest classifier. To set parameters evaluate performance, nested 5-fold cross-validation subject performed. performance our proposed evaluated different multi-class problems. minimum overall accuracy 95.31% 86.64% cross-validation, respectively. promising terms accuracy, sensitivity, specificity compared ones state-of-the-art systems. can be health care applications aim improving classification.

Язык: Английский

---

Automated sleep scoring: A review of the latest approaches

Sleep Medicine Reviews, Год журнала: 2019, Номер 48, С. 101204 - 101204

Опубликована: Авг. 9, 2019

Язык: Английский

---

Automated detection of schizophrenia using nonlinear signal processing methods

Artificial Intelligence in Medicine, Год журнала: 2019, Номер 100, С. 101698 - 101698

Опубликована: Июль 20, 2019

Язык: Английский

---

Investigating the Use of Pretrained Convolutional Neural Network on Cross-Subject and Cross-Dataset EEG Emotion Recognition

Sensors, Год журнала: 2020, Номер 20(7), С. 2034 - 2034

Опубликована: Апрель 4, 2020

The electroencephalogram (EEG) has great attraction in emotion recognition studies due to its resistance deceptive actions of humans. This is one the most significant advantages brain signals comparison visual or speech context. A major challenge EEG-based that EEG recordings exhibit varying distributions for different people as well same person at time instances. nonstationary nature limits accuracy it when subject independency priority. aim this study increase subject-independent by exploiting pretrained state-of-the-art Convolutional Neural Network (CNN) architectures. Unlike similar extract spectral band power features from readings, raw data used our after applying windowing, pre-adjustments and normalization. Removing manual feature extraction training system overcomes risk eliminating hidden helps leverage deep neural network's uncovering unknown features. To improve classification further, a median filter eliminate false detections along prediction interval emotions. method yields mean cross-subject 86.56% 78.34% on Shanghai Jiao Tong University Emotion Dataset (SEED) two three classes, respectively. It also 72.81% Database Analysis using Physiological Signals (DEAP) 81.8% Loughborough Multimodal (LUMED) classes. Furthermore, model been trained SEED dataset was tested with DEAP dataset, which 58.1% across all subjects Results show terms accuracy, proposed approach superior to, par with, reference identified literature limited complexity elimination need extraction.

Язык: Английский

---

Automatic sleep stages classification based on iterative filtering of electroencephalogram signals

Neural Computing and Applications, Год журнала: 2017, Номер 28(10), С. 2959 - 2978

Опубликована: Март 3, 2017

Язык: Английский

---

Intra- and inter-epoch temporal context network (IITNet) using sub-epoch features for automatic sleep scoring on raw single-channel EEG

Biomedical Signal Processing and Control, Год журнала: 2020, Номер 61, С. 102037 - 102037

Опубликована: Июнь 23, 2020

A deep learning model, named IITNet, is proposed to learn intra- and inter-epoch temporal contexts from raw single-channel EEG for automatic sleep scoring. To classify the stage half-minute EEG, called an epoch, experts investigate sleep-related events consider transition rules between found events. Similarly, IITNet extracts representative features at a sub-epoch level by residual neural network captures sequence of via bidirectional LSTM. The performance was investigated three datasets as length (L) increased one ten. achieved comparable with other state-of-the-art results. best accuracy, MF1, Cohen's kappa ($\kappa$) were 83.9%, 77.6%, 0.78 SleepEDF (L=10), 86.5%, 80.7%, 0.80 MASS (L=9), 86.7%, 79.8%, 0.81 SHHS respectively. Even though using four epochs, still comparable. Compared single on average, accuracy MF1 2.48%p 4.90%p F1 N1, N2, REM 16.1%p, 1.50%p, 6.42%p, Above improvement not significant. results support that considering latest two-minute can be reasonable choice scoring networks efficiency reliability. Furthermore, experiments baselines showed introducing intra-epoch context contributes in overall has positive synergy effect learning.

Язык: Английский

---